Hi everyone!
I'm looking for references for standardizing chi-square type goodness of fit
measures (Pearson's, but also generally Cressie and Read's power divergence
statistics).
The reason is I have very very large tables with dfs in the thousands or
more and p-values are generally completely useless for comparing fit in most
of my cases. I've come across some stuff from psychology, e.g. Rasch models
and the like, where it is suggested to either simply divide the statistic
with the df or compute a "z-score" transformation of the statistic thus:
sqrt(2*chi^2) - sqrt(2*df-1) (sic!).
The issue is mainly comparing fit between tables, so I'm not too bothered
about direct interpretation of the value. My p-values right now are all
x^-16, so they are impossible to differentiate between.
I like the idea of simply dividing by df, but I can't seem to find any
references for that or any alternative standardizations?
Also, if anyone knows of any work involving chi-square measures in high
dimensional tables a pointer would be well appreciated!
Thanks!
Maja
You may leave the list at any time by sending the command
SIGNOFF allstat
to [log in to unmask], leaving the subject line blank.
|